* 请先在自己的授权 Stata 中 cd 到完整解压目录，再运行本文件。
version 16.0
clear
set more off
capture log close
log using "Stata课堂运行.log", text replace
* 步骤一：导入并核对四格来源和分母
import delimited using "两州就业汇总.csv", varnames(1) encoding("utf-8") asdouble clear
isid 州别 调查时期
assert _N == 4
assert inlist(州别,"新泽西州","宾夕法尼亚州东部")
assert inlist(调查时期,"政策前","政策后")
assert 有效门店数 > 0 & !missing(有效门店数,全职当量合计)
generate byte treat = (州别 == "新泽西州")
generate byte post = (调查时期 == "政策后")
isid treat post
generate double avgfte = 全职当量合计 / 有效门店数
label variable avgfte "每店平均全职当量"
format avgfte %12.8f
list 州别 调查时期 有效门店数 avgfte, noobs
* 步骤二：依次保存四个均值，再算两次差
quietly summarize avgfte if treat == 1 & post == 0, meanonly
scalar tb = r(mean)
quietly summarize avgfte if treat == 1 & post == 1, meanonly
scalar ta = r(mean)
quietly summarize avgfte if treat == 0 & post == 0, meanonly
scalar cb = r(mean)
quietly summarize avgfte if treat == 0 & post == 1, meanonly
scalar ca = r(mean)
scalar treated_change = ta - tb
scalar control_change = ca - cb
scalar counterfactual = tb + control_change
scalar did = treated_change - control_change
display "处理变化 = " %12.8f treated_change
display "对照变化 = " %12.8f control_change
display "DID = " %12.8f did
assert abs(did-2.75360578298058) < 0.00000001
* 步骤三：显式设置基组，交互项只核对点估计
regress avgfte ib0.treat##ib0.post
display "交互项 = " %12.8f _b[1.treat#1.post]
assert abs(_b[1.treat#1.post]-did) < 0.00000001
assert e(df_r) == 0
preserve
rename treat 处理组
rename post 政策后
rename avgfte 每店平均全职当量
export delimited using "Stata四格均值结果.csv", replace
restore
* 步骤四：读入教学反事实情景，不覆盖真实调查
preserve
import delimited using "反事实情景.csv", varnames(1) encoding("utf-8") asdouble clear
assert _N == 1 & !missing(假设对照组变化)
scalar assumed_change = 假设对照组变化[1]
scalar scenario_did = treated_change-assumed_change
scalar scenario_counterfactual = tb+assumed_change
display "假设情景DID = " %12.8f scenario_did
display "假设情景反事实 = " %12.8f scenario_counterfactual
restore
* 步骤五：用新的原创模拟门店资料独立迁移
import delimited using "独立练习_模拟门店.csv", varnames(1) encoding("utf-8") asdouble clear
isid 门店编号 时期
assert _N == 16
assert inlist(组别,"试点组","参照组") & inlist(时期,"政策前","政策后")
bysort 门店编号: assert _N == 2
bysort 门店编号: assert 组别 == 组别[1]
generate byte treat = (组别 == "试点组")
generate byte post = (时期 == "政策后")
collapse (mean) avgfte=全职当量 (count) stores=全职当量, by(treat post)
isid treat post
assert _N == 4 & stores == 4
list treat post stores avgfte, noobs
* 步骤六：模拟的四格回归只用于检查运算
regress avgfte ib0.treat##ib0.post
scalar simulated_did = _b[1.treat#1.post]
assert abs(simulated_did-3) < 0.00000001
assert e(df_r) == 0
display "模拟DID = " %9.4f simulated_did
rename treat 处理组
rename post 政策后
rename avgfte 每店平均全职当量
rename stores 门店数
export delimited using "Stata模拟四格结果.csv", replace
log close